MARS workup Β· 2026-07-04
A pure-private venture, a public-market alt-asset behemoth, and one that just crossed over. Same shape, same joins, same substrate β shown against the tables MARS has been building.
MARS's job is visibility into companies the public market hasn't seen yet. The harder questions are continuity β does the substrate carry a company cleanly across the private-to-public crossover? β and scale β does it hold up on a firm that does dozens of billion-dollar deals a year? The three cases below answer both. Every number came from a single SELECT against mars.companies_v2 joined through the v2 canonical layer. No hand entry.
01 The private-only case
Twelve years of SEC Reg D substrate on a $60B+ private company β visible only because we ingest Form D directly.
Directly from mars.form_d_filings_v2
| Filed | Offering size | Sold | Round it funded |
|---|---|---|---|
| 2013-09-25 | $15 M | $14 M | Series A |
| 2014-07-03 | $33.7 M | $33.4 M | Series B |
| 2017-08-22 | $140 M | $140 M | Series D |
| 2019-02-07 | $250 M | $250 M | Series E |
| 2019-11-05 | $400 M | $400 M | Series F |
| 2021-02-02 | $1.0 B | $1.0 B | Series G |
| 2021-09-15 | $1.6 B | $1.6 B | Series H |
| 2023-09-29 | $1.22 B | $1.18 B | Series I (dual filing) |
| 2023-11-13 | $1.0 B | $685 M | Series I follow-on |
| 2025-01-02 | $10.0 B | $8.59 B | Series J mega round |
| 2025-09-23 | $1.74 B | $1.62 B | Series K (dual) |
| 2025-12-31 | $5.0 B | $4.08 B | Series L |
Twelve of eighteen shown. The whole file is WHERE company_id = '32c37β¦' β same UUID resolves every Form D filing against the same canonical company record.
JSONB investor arrays canonicalized to investors_v2
| Date | Round | Size | Investors |
|---|---|---|---|
| 2026-02-09 | Equity + Debt | $5.0 B | Goldman Sachs, Morgan Stanley, JPMorgan, Neuberger Berman, Qatar Investment Authority, Glade Brook Capital |
| 2025-12-16 | Series L | $4.0 B | Andreessen Horowitz, Insight Partners, Fidelity, JPMorgan AM, BlackRock, Blackstone, Coatue, GIC, MGX, NEA, Ontario Teachers, Robinhood Ventures, T. Rowe Price, Temasek, Thrive Capital, Winslow Capital |
| 2025-09-09 | Series K | $1.0 B | Andreessen Horowitz, Insight Partners, MGX, Thrive Capital, WCM Investment Management |
Same company_id on the acquirer side of merger_deals_v2
| Announced | Target | Value | Note |
|---|---|---|---|
| 2026-06-16 | Panther Labs | undisc. | Cloud security analytics |
| 2025-05-14 | Neon | $1.0 B | Serverless Postgres β vs Snowflake Iceberg push |
| 2025-02-04 | BladeBridge | undisc. | Legacy-warehouse migration |
| 2024-06-04 | Tabular | $1.75 B | Iceberg format co-creator; bought hours after Snowflake talks broke down |
| 2023-10-23 | Arcion | $100 M | Real-time replication |
| 2023-07-19 | MosaicML | $1.3 B | Generative-AI training platform; launched Databricks's foundation-model story |
| 2023-05-03 | Okera | undisc. | Governance / access control |
Named on Form D filings across the twelve years above and resolved through form_d_officers_v2 to canonical persons_v2 rows.
| Name | Roles | Note |
|---|---|---|
| Ali Ghodsi | Executive Officer Β· Director Β· Promoter | CEO |
| Ion Stoica | Executive Officer Β· Director Β· Promoter | Co-founder Β· Berkeley AMPLab |
| Matei Zaharia | Executive Officer Β· Director Β· Promoter | Co-founder Β· created Spark |
| Ben Horowitz | Director | Andreessen Horowitz |
| David Conte | Executive Officer | CFO |
| Elena Donio | Director | Board |
| Jonathan Chadwick | Director | Board Β· former VMware CFO |
Sample of 142 officer rows resolved to canonical persons. Extractor typos β Ali Ghosdi for Ali Ghodsi β get queued through the Qwen β Haiku β Grok-web resolution tier chain and cleared without human intervention. Same person_id on either side of the merge.
02 The scale case
The world's largest real-estate investor doing dozens of billion-dollar deals across property, credit, and M&A β all threaded together by the reit vertical MARS scaffolded two days ago.
From real_estate_deals_v2 Β· reit vertical, live since 2026-07-03
| Type | Count | Total |
|---|---|---|
| Property acquisition | 19 | $96.5 B |
| Financing | 9 | $74.9 B |
| REIT offering | 4 | $42.8 B |
| Fund raise | 3 | $22.1 B |
| Property sale | 16 | $15.5 B |
| Development JV | 2 | $10.0 B |
| Portfolio sale | 1 | $1.0 B |
| Date | Type Β· asset | Value |
|---|---|---|
| 2026-06-03 | Industrial acquisition | $81.0 B |
| 2026-06-01 | Residential financing | $60.0 B |
| 2026-05-05 | Data-center REIT offering | $16.0 B |
| 2026-04-14 | Data-center REIT offering | $16.0 B |
| 2026-03-27 | Fund raise | $15.0 B |
| 2026-01-22 | Data-center acquisition | $10.0 B |
The concentration in data centers is real β Blackstone owns QTS Realty, and the AirTrunk $16.1B acquisition (below) added Asia-Pacific footprint. Data-center deals cluster in real_estate_deals_v2 as property_type='data_center'; the M&A shows up separately in merger_deals_v2. Same company_id stitches both.
Where the RE strategy crosses into operating-company M&A
| Announced | Target | Value | Sector |
|---|---|---|---|
| 2024-09-04 | AirTrunk | $16.1 B | Data centers Β· Asia-Pacific |
| 2022-10-31 | Copeland (formerly Emerson Climate) | $14.0 B | HVAC / industrial |
| 2024-04-08 | Apartment Income REIT (AIR Communities) | $10.0 B | Residential |
| 2024-04-29 | L'Occitane International S.A. | $6.4 B | Consumer |
| 2025-02-24 | Safe Harbor Marinas | $5.65 B | Marina operator |
| 2023-03-14 | Cvent Holding Corp. | $4.6 B | Event tech |
63 M&A rows total. The three duplicate rows on Apartment Income REIT β Apartment Income REIT Corp. / Apartment Income REIT (AIR Communities) / Apartment Income REIT Corp. (AIR Communities) β are exactly what the entity-resolution tier chain is built to collapse. Next Phase G fire, Qwen resolves them; if it can't, Grok-web does.
03 The private-to-public arc
Same substrate carries the whole trajectory: 2019 seed Form D to 2025 IPO to $3.28B of insider Form 144 filings, all under one company_id.
Selected billion-plus rounds from funding_deals_v2
| Date | Round | Amount | Notable investors |
|---|---|---|---|
| 2026-05-19 | Debt + Equity | $20.0 B | NVIDIA Strategic |
| 2026-05-18 | Delayed-Draw Term Loan | $3.1 B | Morgan Stanley, MUFG |
| 2026-04-09 | Convertible Senior Notes | $3.25 B | Public offering |
| 2026-03-31 | Debt Facility | $8.5 B | Goldman Sachs, JPMorgan, Morgan Stanley, MUFG |
| 2025-03-20 | IPO | $2.41 B | NASDAQ:CRWV |
| 2024-03-31 | Debt Facility (pre-IPO) | $8.0 B | MUFG, Morgan Stanley, Goldman, JPMorgan |
| Announced | Target | Value | Note |
|---|---|---|---|
| 2025-10-30 | Marimo Inc. | undisc. | AI-notebook workspace |
| 2025-10-06 | Monolith AI | undisc. | UK physics-ML startup |
| 2025-09-03 | OpenPipe Inc. | undisc. | Fine-tuning platform |
| 2025-07-07 | Core Scientific | $9.0 B | Public-to-public; Bitcoin miner β AI DC footprint |
| 2025-05-05 | Weights & Biases | undisc. | MLOps platform |
| 2023-01-25 | Conductor Technologies | undisc. | Render farm; pre-IPO tuck-in |
Top sellers from form_144_v2 Β· resolved to persons_v2
| Seller | Role at issuer | Filings | Value |
|---|---|---|---|
| West Clay Capital LLC | Brannin McBee vehicle | 32 | $938.5 M |
| Jack Cogen | Director | 7 | $715.5 M |
| Brannin McBee | Co-founder, Executive Officer | 27 | $395.2 M |
| Michael Intrator | Co-founder, CEO, Director | 28 | $266.4 M |
| Brannin J. McBee 2022 IRR Trust | McBee family trust | 7 | $258.2 M |
| Total across all 335 filings | 335 | $3.28 B |
Michael Intrator and Brannin McBee both appear in form_d_officers_v2 from 2019 onward. The person_id on those officer rows is the same UUID resolved on the Form 144 sale rows β cross-linked at ingest, not stitched together at query time.
04 The through-line
The value isn't the count of rows. It's that everything joins cleanly.
Both companies live under a single companies_v2.company_id. Fourteen v2 tables carry FKs to it: funding, mergers, real estate, four SEC substrates, two officer bridges, deal articles, canonical persons, and the reit vertical's real_estate_deals_v2.
The join graph is discovered at runtime from Postgres foreign keys β new tables get picked up without code changes.
Form D, 13D, 13G, 144, and ADV all land in v2 tables owned by mars. Officer parsing runs at ingest time, not on read. That's how Databricks has 142 canonical officer rows without a public 10-K, and how CoreWeave's pre-IPO officers are the same rows as its post-IPO insider sellers.
Every deal carries an integrity_tier: verified / vouched / warn. Consumer queries filter to the first two; the small "warn" tail is quarantined and auto re-audited. The $10B Databricks Form D and the $9B CoreWeave / Core Scientific merger both cleared the gate on ingest.
Extractor typos like Ali Ghosdi, Biran Venturo, Michael Intrrator are auto-resolved through Qwen β Haiku β Grok-web. 891 canonical merges executed 2026-07-04 with 0 unique-constraint conflicts. The person named on a 2013 Form D stays the same person named on a 2026 Form 144.